Skip to content

What can the RTX 5070 run?

What fits on the RTX 5070
Text models: 246Visual models: 12Strongest first
Memory
12 GB
Bandwidth
672 GB/s
Maker
NVIDIA

Open the calculator on this machineEvery model for 12 GB of VRAM

Text models that run on it

Text models that run on the RTX 5070, best first
ModelBest fitContextSpeedStrongest at
Qwen3.5 9B8-bit24K49 tok/sChat, Coding, Agents, Docs
Gemma 4 12B6-bit112K45 tok/sChat, Coding, Docs
Qwen3.5 4B8-bit176K86 tok/sAgents, Docs
Ministral 3 14B4-bit16K55 tok/sCoding, Agents, Docs
Gemma 4 E4B8-bit128K77 tok/sChat, Coding, Docs
gpt-oss 20B3-bit40K193 tok/sAgents
Ministral 3 8B6-bit28K64 tok/sDocs
Gemma 4 E2B8-bit128K137 tok/sChat
Nemotron 3 Nano 4B8-bit256K106 tok/sDocs
Llama 3.1 8B8-bit20K56 tok/sChat, Coding, Agents, Docs
Qwen3 0.6B8-bit40K394 tok/sNot rated
Qwen3 14B3-bit24K64 tok/sNot rated
Qwen3 VL 8B Instruct6-bit24K65 tok/sText and images
Qwen3 8B8-bit16K55 tok/sNot rated
Qwen2.5 7B Instruct8-bit32K59 tok/sNot rated
Qwen2.5 VL 7B Instruct8-bit40K55 tok/sText and images
Unlimited OCR8-bit32K303 tok/sReading documents
Qwen2.5 1.5B Instruct8-bit32K231 tok/sNot rated
Qwen3.5 2B8-bit256K170 tok/sText and images
Qwen3 VL 4B Instruct8-bit40K90 tok/sText and images
Qwen3 1.7B8-bit40K184 tok/sNot rated
Qwen3 4B8-bit40K98 tok/sNot rated
Qwen3 4B Instruct 25078-bit44K98 tok/sNot rated
Qwen2.5 3B Instruct8-bit32K124 tok/sNot rated
DeepSeek OCR8-bit8K303 tok/sReading documents
Qwen2.5 Coder 14B Instruct3-bit20K63 tok/sCoding
Qwen2.5 VL 3B Instruct8-bit125K105 tok/sText and images
Qwen2.5 Coder 7B Instruct8-bit32K59 tok/sCoding
Mistral 7B Instruct V0.28-bit28K59 tok/sNot rated
Chandra OCR 28-bit160K77 tok/sReading documents
Qwen3 VL 2B Instruct8-bit72K177 tok/sText and images
Qwen2 VL 2B Instruct8-bit32K172 tok/sText and images
MiniCPM5 2B8-bit128K158 tok/sNot rated
Gemma 3 4B IT8-bit128K109 tok/sText and images
Llama 3.2 3B Instruct8-bit64K124 tok/sNot rated
LightOnOCR 2 1B8-bit16K321 tok/sText and images
Qwen2.5 14B Instruct3-bit20K63 tok/sNot rated
Qwen3 1.7B Base8-bit32K212 tok/sNot rated
HunyuanOCR8-bit96K308 tok/sText and images
Qwen3 0.6B Base8-bit32K457 tok/sNot rated
Rax 4.58-bit256K170 tok/sText and images
Qwen2 VL 7B Instruct8-bit32K55 tok/sText and images
Ornith 1.5 9B8-bit20K49 tok/sNot rated
Unlimited OCR 28-bit120K302 tok/sReading documents
Granite 4.1 3B8-bit88K113 tok/sNot rated
DeepSeek R1 Distill Qwen 1.5B8-bit128K231 tok/sNot rated
Mistral 7B V0.18-bit32K59 tok/sNot rated
Qwen2.5 Coder 3B Instruct8-bit32K124 tok/sCoding
Meta Llama 3 8B Instruct Abliterated V38-bit8K56 tok/sNot rated
EXAONE 3.5 7.8B Instruct8-bit20K57 tok/sNot rated
Saiga Llama3 8B8-bit8K56 tok/sNot rated
Qwen2.5 Coder 1.5B Instruct8-bit32K231 tok/sCoding
DeepSeek R1 0528 Qwen3 8B8-bit16K55 tok/sNot rated
Gemma 3 12B IT5-bit28K55 tok/sText and images
Phi 3.5 Mini Instruct8-bit16K107 tok/sNot rated
Qwen3 4B Thinking 25078-bit44K98 tok/sNot rated
Phi 4 Mini Instruct8-bit52K104 tok/sNot rated
Dots.ocr8-bit128K139 tok/sReading documents
DeepSeek R1 Distill Qwen 7B8-bit52K59 tok/sNot rated
Phi 3 Mini 4K Instruct8-bit4K107 tok/sNot rated

Best fit is the highest quality, up to 8-bit, that leaves the use case’s context. Ranked by our use-case scores, then by Hugging Face downloads. Speeds are rough, plus or minus 25%. The top 60 of 246 are shown.

Image and video models

Image and video models that run on the RTX 5070
ModelBest fitPer outputStrongest at
Stable Diffusion XL 1.016-bit11 sImages
Z-Image Turbo8-bit15 sImages
FLUX.1 [dev]6-bit34 sImages
Qwen-Image 2.18-bit4.0 minImages
FLUX.3 Klein16-bit22 sImages
FLUX.2 [klein] 9B8-bit7.7 sImages
Krea 2 Turbo5-bit34 sImages
Stable Diffusion 3.5 Large8-bit2.1 minImages
MOSS-Image16-bit2.6 minImages
Wan 2.2 TI2V-5B16-bit23 minVideo
MOSS Image Base16-bit2.6 minImages
HunyuanVideo 1.56-bit1.9 hVideo

Only just too big

Questions

What is the best LLM for the RTX 5070?

Of the open models we track, the strongest that run well on the RTX 5070 are Qwen3.5 9B, Gemma 4 12B, Qwen3.5 4B. 246 text models fit it in all.

Calculated from published specs, not benchmarked. All machines